Recommendation Approaches

Recommendation Systems Approaches

The idea of a recommendation system is to predict the future preferences of a consumer based on their recorded preferences of the past as well as any other measurable information about the customer. Two primary approaches have been used in the past.

First, collaborative filtering involves using information about the preferences of the consumer to infer the items they will like next. Item-to-item collaborative filtering has been used at as part of their product recommendation services. Similarly, Facebook and LinkedIn have used the same approach to recommend music, friends, and colleagues you may be interested in.

Second, content-based filtering is based on analyzing product or service descriptions and stated preferences (e.g. think of a consumer pre-selecting categories they are interested in) to co-predict the types of future items that a user would like. This is much more like the regression and classification models you generated earlier in the course. Content-based filtering has been used by Rotten Tomatoes and Pandora.

Naturally, as you might have guessed, the best recommendation systems use a combination of both techniques. Netflix uses a hybrid approach that is based on both collaborative filtering (i.e. which movies you've watched in the past) and content-based filtering (your stated preferences).

You will learn how to build a hybrid approach in Azure ML Studio through the tutorial's in this chapter.